Comparative Assessment of Machine Learning Models for Drilling Parameter Optimization in Offshore Wells

This research delves into optimising offshore drilling by leveraging sophisticated machine learning methodologies, with a particular emphasis on forecasting and scrutinising the Rate of Penetration (ROP) within the Volve Oil Field, North Sea. Since offshore drilling’s inception in the late 19th century, it has cemented itself as a cornerstone of global energy supply, continually evolving through technological breakthroughs that bolster both its efficacy and safety. Addressing a pivotal challenge, this study concentrates on refining ROP prediction frameworks to streamline drilling activities, curtail operational expenditures, and mitigate ecological repercussions. The research’s importance is underscored by its exhaustive exploration of drilling variables and their intricate interplay with ROP, integrating real-time drilling data alongside Computed Petrophysical Output (CPO) log records. Through machine learning-driven methodologies, the study aspires to construct robust predictive models capable of augmenting drilling efficiency and refining strategic decision-making in offshore ventures. Ultimately, this work extends the existing repository of knowledge on drilling optimisation while illustrating the tangible benefits of machine learning applications within the oil and gas sector. The methodological framework encompassed an in-depth analysis of drilling data leveraging three distinct machine learning paradigms: Decision Tree Regression, XGBoost Regression, and Histogram Gradient Boosting Regression. The study meticulously explored interdependencies among multiple drilling variables, such as Weight on Bit (WOB) and Surface Rotation Per Minute (SURF_RPM), alongside geological attributes like porosity (PHIF) and shale volume (VSH). To gauge predictive efficacy, the models were assessed through R-squared values, Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Empirical results underscored XGBoost Regression as the most consistent performer, yielding robust predictive accuracy across both training (𝑅 2 = 0.889) and testing (𝑅 2 = 0.777) datasets. Notably, the investigation uncovered substantial correlations among drilling parameters, including a pronounced inverse relationship between water saturation (SW) and shale volume (VSH). The study inferred that machine learning methodologies, particularly XGBoost, establish a dependable framework for forecasting the Rate of Penetration (ROP) in offshore drilling scenarios. These insights pave the way for real-time drilling adjustments, enhancing operational efficiency while aligning with environmentally responsible practices in offshore resource extraction